3.3 Network-Attached Storage
Like the acronym SAN, NAS is largely a marketing term that has, through repeated use, gained a technical definition. The primary distinction between NAS and SAN rests on the difference between data files and data blocks. NAS transports files; SANs transport blocks. NAS uses file-oriented delivery protocols such as NFS and CIFS, whereas SANs use block-oriented delivery protocols such as SCSI-3. Because data blocks are the raw material from which files are formed, NAS also has a block component. These blocks are addressed on a per-file basis, using meta-data (directory information) to determine which file to use. The block access methods of a NAS device, however, are typically hidden in the NAS enclosure. To the outside world, the NAS device is a server of files and directories.
NAS accomplishes the central goal of storage networking: the sharing of storage resources through the separation of servers and storage over a common network. Like SAN-based storage, NAS overcomes the limitations of parallel SCSI and enables a more flexible deployment of shared storage. In redundant configurations, NAS can also provide highly available, nondisruptive storage access.
As shown in Figure 37, a NAS architecture includes disk arrays for data placement, a NAS processor, and an external interface to the user network. This architecture has a number of implications. Although the block data transport of a SAN normally occurs over a dedicated storage network (or VLANs within a Gigabit Ethernet network), the file transport of a NAS device assumes direct connectivity to the user network. NAS performance is thus partially determined by the bandwidth available on the messaging network. In addition, although the NAS processor is optimized for file transport, it is essentially a thin server sitting on a LAN "front-ending" storage arrays. To avoid the definition of NAS as "network-attached server" or simply "file server," vendors of NAS products have attempted to make the thin server component so thin that it is invisible from the user's perspective. The term NAS is therefore always linked in marketing literature to the concept of "appliance"a device that is simply plugged into the network and requires little administration. Finally, the connection between NAS intelligence and its disks is immaterial from a user's perspective. Although SAN storage is predicated on high-performance gigabit interfaces for storage (either directly or via bridge products), a NAS device may rely on parallel SCSI, IDE, or Fibre Channel for storage connectivity.
Figure 37 NAS architecture.
The challenge of NAS vendors is to make their products more appliancelike and to reduce the overhead of NFS/CIFS protocols over TCP/IP. Although TCP/IP imposes its own latency on file transactions, NFS and CIFS also engender latency as file transfer sessions are established and torn down, and as files are found through directory lookup. Chip-based TOEs offer some relief, although the main initiative from some NAS vendors for dealing with latency is support of the Direct Access File System (DAFS) over the VI protocol. DAFS relies on remote direct memory access (RDMA) techniques to move data directly to systems memory. If a NAS controller can place file data directly into the memory of the client efficiently, transport latency is offset by the higher performance of file data placement.
The convergence of NAS and SAN is accelerated by the development of IP storage networks. With 10Gb backbones and gigabit interfaces to end devices, both file and block data can share a common network infrastructure. This promotes the deployment of shared storage solutions on the basis of user application requirements, without the artificial limits imposed by incompatible network topologies.